Davidsonian Semantics and Anaphoric Deflationism

Transcription

1 Davidsonian Semantics and Anaphoric Deflationism David Löwenstein Abstract Whether or not deflationism is compatible with truth-conditional theories of meaning has often been discussed in very broad terms. This paper only focuses on Davidsonian semantics and Brandom s anaphoric deflationism and defends the claim that these are perfectly compatible. Critics of this view have voiced several objections, the most prominent of which claims that it involves an unacceptable form of circularity. The paper discusses how this general objection applies to the case of anaphoric deflationism and Davidsonian semantics and evaluates different ways of responding to it (Williams 1999, Horisk 2008 and Lance 1997). Then, three further objections to the compatibility of these theories are assessed and eventually dismissed (Horisk 2007, Patterson 2005 and Collins 2002). It is shown how these considerations shed light on core issues of the debate. Keywords: Robert Brandom, Donald Davidson, Michael Williams, anaphora, circularity, deflationism, meaning, prosentence, prosententialism, semantics, truth, truth-conditional theory of meaning. Final Draft, published in 2012 in Dialectica 66. The definitive version is available at Freie Universität Berlin. david.loewenstein@fu-berlin.de. 1

2 Contents 1 Introduction 2 2 Preliminaries Anaphoric Deflationism Davidsonian Semantics Compatibility Prosentences in T-sentences Expression and Explanation Truth in Radical Interpretation Circularity Why the Circularity Objection Misfires Other Proposals Compatibility Defended Expressive Inadequacy The Metalanguage Objection Cognitive Opacity and the Unity of Truth Conclusion 29 References 30 1 Introduction It is widely held that the concept of truth plays an indispensible explanatory role in theories of meaning. Deflationists, by contrast, argue that this concept is merely an expressive device, lacking any explanatory power. On these grounds, it has become commonplace to say that deflationism and truthconditional approaches to meaning are mutually incompatible. 2

3 This topic has often been discussed in very broad terms, encompassing many varieties of deflationism and many truth-conditional theories of meaning. The scope of this paper is comparatively modest. I will only focus on the compatibility of one truth-conditional approach to meaning, Donald Davidson s, with one species of deflationism, Robert Brandom s prosententialism or anaphoric deflationism. I will argue that the widespread view that these theories are incompatible a view shared by Davidson and Brandom themselves is an illusion fueled by mutual misconceptions of these theories. To this end, I will start by spelling out what I take anaphoric deflationism and Davidsonian semantics to consist in (section 2). Second, I will try to establish their compatibility by holding Davidsonian semantics fixed and showing how anaphoric deflationism can account for the role of the concept of truth in this approach (section 3). While this claim has been supported, among others, by Michael Williams (1999) and Max Kölbel (2001), its critics have voiced several serious objections. In section 4, I will address the most prominent among these concerns, which is also shared by Brandom and Davidson, namely the idea that combining deflationism with truth-conditional approaches to meaning involves an unacceptable form of circularity. After discussing how this argument applies to the present case, I will comment on different responses to it by Michael Williams (1999), Claire Horisk (2008) and Mark Lance (1997). Finally, in section 5, I will assess and eventually dismiss three further objections by Claire Horisk (2007), Douglas Patterson (2005), and John Collins (2002). These considerations will shed light on core issues of the debate and clarify both Davidsonian semantics and anaphoric deflationism, as well as the way they are compatible. 2 Preliminaries In this preliminary section, I will briefly characterize both anaphoric deflationism and Davidsonian semantics. But I will not discuss independent 3

4 problems about either of these. 1 My concern is, assuming both theories are prima facie admissible, whether or not they are compatible. 2.1 Anaphoric Deflationism While there is some disagreement over how best to define deflationist conceptions of truth, 2 I take it that the most fruitful proposal relies on distinguishing two questions (compare Armour-Garb & Beall 2005b). The substantial question What is truth? The functional question What is the conceptual role of truth and the linguistic role of true and related expressions? Traditionally, philosophers take both questions seriously. They give an account of truth, and then employ this account in order to explain its role in thought and talk. I will call these proposals substantial theories of truth or substantivism. By contrast, deflationists think that asking what truth is is at least uninteresting and maybe even deeply misguided. They hold that answering the functional question is the only interesting thing to do. In shrugging their shoulders at the substantial question, deflationists are local metaphysical quietists: As for the alleged nature of truth, they do not undertake any ontological commitment whatsoever. Deflationary theories differ in how to make sense of the functional role of truth. As its name already reveals, anaphoric deflationism 3 relies on anaphoric mechanisms. That is, it relies on relations between linguistic expressions that allow some expression tokenings, anaphoric dependents, to 1 For example, I will leave out independent worries concerning anaphoric deflationism. Compare, for example, Wilson 1990 and Båve 2009, as well as footnotes 5, 6 and 8. 2 Deflationism about truth is often linked to deflationism about reference. This characterization of deflationism easily translates to this and other cases. 3 For the original presentation and development of the theory, compare Grover et al. 1975, Brandom 1988, 1994 and Note also that anaphoric deflationism about truth has a sibling in anaphoric deflationism about reference, as developed in Brandom 1984 and These positions are related, but, strictly speaking, independent. 4

5 anaphorically inherit the meanings of other expression tokenings, anaphoric antecedents. Pronouns are paradigmatic examples for anaphoric expressions: (1) Even though a friend of mine strongly believes in him, she could never convince me to believe in God myself. The tokening of she is anaphorically dependent on its antecedent a friend of mine. It inherits its meaning by referring back to that antecedent. Also, the tokening of him is dependent on its antecedent God. In this case, the it refers forth to that antecedent and thereby inherits its meaning. 4 Anaphoric deflationism treats complex utterances formed with expressions like is true as prosentences. Hence, the theory is also called prosententialism. Prosentences are anaphorically dependent sentences that inherit their meanings from other sentence tokenings. Truth talk, the claim goes, does not involve attributions of a property but the employment of a unique anaphoric prosentence-forming device the truth operator. 5 Consider the following example of ordinary truth talk. (2) Sarah s self-description is true. Understood as a prosentence, (2) inherits its meaning from a claim by Sarah, a sentence tokening picked out by the expression Sarah s self-description. Thus, in a context where Sarah has said I am a maverick, (2) means: (3) Sarah is a maverick. 4 Linguists call the latter phenomenon cataphora and reserve the term anaphora for backwards cases like the former. Cataphora and anaphora are, in turn, subclasses of endophora. Thus, strictly spreaking, the theory is endophoric rather than anaphoric. I will nevertheless stick to the established term anaphoric deflationism. 5 Prosententialism holds that truth talk is not essentially tied to the expression is true, but to the linguistic function of forming prosentences. Since Brandom understands facts as true sentences (e.g. in Brandom 1994, 327), he is committed to say that is a fact is an instance of the truth operator, too. However, he never explicitly defines the truth operator, which would then pick out all and only its instances. This is problematic because he relies heavily on distinguishing the anaphoric truth operator from the standard idea of a truth predicate (in Brandom 1994, 1997 and 2002). In Löwenstein 2010, I have proposed a definiton of the truth operator which solves this problem. 5

6 Other instances of truth talk add a quantificational dimension to the story. For example, (4) is understood as (5) and (6) as (7). (4) Some of these provocative remarks are actually true. (5) There are some sentence tokenings t such that (i) t is among these provocative remarks, and (ii) actually, t. (6) Everything John just said is true. (7) For all sentence tokenings t, if John just said t, then t. Given the occurence of the variables both inside and outside of quotation marks, these quantifiers have to be understood substitutionally. 6 Prosentences are generic in that any sentence tokening can be the antecedent of a prosentence, given that it is specified uniquely, e.g. by describing or quoting it. 7 Prosentences inherit their meanings from the set of their anaphoric antecedents a singleton in lazy cases like (2) as opposed to quantificational cases like (4) and (6). And prosentences can occur freestanding or embedded in logically complex sentences, e.g. as antecedents of conditionals. Given these resources, the core prosententialist claim is that every use of true can be explained in these terms. 8 Prosentences enhance the expressive power of a language. For example, they allow for rejecting one from a set of claims without specifying which, like in (8), and for endorsing large or even infinite sets of claims, as in (9). 6 Prosententialism therefore presupposes an apt account of substitutional quantification, including a treatment of sentences involving indexicals, context-sensitivity in general and other challenges. Here, however, I cannot discuss, but only point out this commitment. 7 Prosentences which specify their antecedents by quotation still inherit their meanings from sentence tokenings rather than from sentence types. However, it is possible to make sense of sentence types in terms of sentence tokenings. In section 3.1, I show how. 8 This implies that the Liar is a prima facie admissible prosentence. While Grover (1977) has attempted a prosententialist dissolution of the Liar, I cannot discuss this problem here. Further, prosententialism also applies to uses of true which are prima facie non-anaphoric as well as to prosentences with antecedents that go forever unuttered, maybe even necessarily so. Even granting Brandom s use of anaphora where he means endophora (compare footnote 4), it remains to be discussed how these cases can be dealt with. 6

7 (8) At least one of Brandom s claims is false. (9) All theorems of first-order logic are true. Further, prosentences can be used in order to endorse or reject claims without knowing what these come down to. Examples of such blind endorsements, as they have come to be called, involve future antecedents like in (10) as well as utterances one has heard, but not understood (compare section 5.3). (10) What the editorial of tomorrow s Times will say is true. Finally, anaphoric deflationism also implies an important scheme: Truth Scheme φ is true just in case φ. Unsurprisingly, the left hand side of this biconditional is interpreted as a prosentence. Thus, if the claim quoted on the left is the one used on the right, they will share their meaning by anaphoric inheritance and thereby validate the biconditional. More generally, the scheme holds whenever the claim quoted on the left is synonymous with the one used on the right. Their identity is just the most trivial way for them to be synonymous. 2.2 Davidsonian Semantics Let me now turn to Donald Davidson s approach to meaning. While there is a lot to say about the intricacies of Davidsonian semantics, I will content myself with a rough sketch and focus on the present issue: the role of the concept of truth. 9 But some aspects will be spelled out in more detail later. Davidsonian semantics builds on Tarski s theory of formalized languages (1944) in order to characterize the meanings expressed in some object language L. Such theories should enable us to derive adequate instances of Tarski s famous Convention T. 9 I mainly rely on Davidson s 1967, 1973, 1974, 1982, 1986 and 1991 articles. 7

8 Convention T s is true if and only if p By replacing s with a sentence quoted from the object language and p with a translation of s in the metalanguage, we arrive at instances of this scheme T-sentences such as: (11) Hunde bellen is true if and only if dogs bark. Tarski used the T-sentences in order to explain the concept of truth, relying on the meanings of the object language sentences they quote. For him, truth in a given object language can be defined as whatever fulfills this function for all T-sentences. Davidson, by contrast, turns Tarski s explanatory strategy upside down: He uses the T-sentences in order to explain the meanings of the object language sentences they quote, relying on the concept of truth. In the T-sentences, he claims, the meanings of the quoted sentences are given on the right hand side, in what is commonly called a truth-condition. Davidson holds that T-sentences are empirical hypotheses about the meanings of object language sentences. But what counts as evidence for or against a given T-sentence? The answer to this question is Davidson s elaborate theory of radical interpretation that is, of the interpretation of the linguistic behavior of speakers from scratch. In this theory, the principle of charity plays a crucial role: If the interpretation of linguistic behavior is to be possible at all, the interpreter has to understand the interpretee as rational and reasonable as far as possible. This entails that what the interpretee says must be understood as as logically coherent and as true wherever possible. Thus, interpreters must be able to distinguish between what the interpretee believes or holds true and what actually is true. Radical interpretation involves a basic distinction between mere belief and actual truth. The principle of charity is crucial because the interpreter has to determine both the meanings of utterances on the basis of their speaker s beliefs and their beliefs on the basis of the meanings of their utterances. Thus, the very possibility of interpreting and eventually understanding a speaker depends 8

9 on a way to break into this circle that is, on charity. The underlying picture of this argument is what Davidson calls triangulation : The interpreter understands the interpretee as causally situated in the very world she herself is causally situated in. In canonical situations, she can assume that the interpretee s utterances concern the very objects she perceives herself. This enables her to arrive at first tentative hypotheses, first T-sentences, which can be used as an entering wedge into interpreting the language as a whole. 3 Compatibility Having set the stage, I will now go on to show why anaphoric deflationism is compatible with Davidsonian semantics. Starting from my rough characterization of Davidsonian semantics, I will go through the points where the concept of truth occurs in this theory in the T-sentences themselves (sections 3.1 and 3.2) and in the theory of radical interpretation (section 3.3). 3.1 Prosentences in T-sentences I think that prosententialism makes perfect sense of the role of truth in the T-sentences. For example, the left hand side of (11) is understood as an embedded prosentence which inherits its meaning from an object language utterance of Hunde bellen. The whole T-sentence (11) asserts the equivalence of this prosentence with the sentence used on the right hand side, Dogs bark. Thus, if (11) is true, Dogs bark is equivalent not only with the prosentence, but also with its anaphoric antecedent, the object language utterance of Hunde bellen. This is precisely what T-sentences are supposed to do in Davidsonian semantics. In the words of Michael Williams, a prosentence is an ideal candidate for this job (1999, 557). Really? While Davidsonian semantics is intended as a theory of meaning for sentence types rather than individual sentence tokenings, prosentences inherit their meanings from sentence tokenings rather than types. And the 9

10 T-sentences are supposed to follow from a finite set of axioms on the basis of the structure of the object language sentences, while prosentences inherit their meanings from sentence tokenings as wholes. However, both of these problems can be solved within the prosententialist setting. In order to see how, recall first that prosentences inherit their meanings from the set of their antecedents, and consider second that sentence types can be understood as sets of sentence tokenings. Thus, all a prosentential reading of the T-sentences requires is an order of explanation that starts with individual sentence tokenings and deals with the question of their possible type-identity later. But this is just the order of explanation radical interpreters have to pursue anyway for reasons independent from prosententialism. At the outset, when looking at foreign utterances in canonical situations, it is an open question whether several such utterances eventually turn out to be equivalent whether they turn out to be tokenings of the same type. Interpreters have to stipulate which ones are tokenings of the same type which utterances are equivalent and test these stipulations empirically. 10 This process is already built into the practice of radical interpretation. When I use a T-sentence in order to interpret some utterance, I presuppose that that utterance is an utterance of the sentence quoted in the T-sentence that the tokening I interpret is a tokening of the sentence type for which the T-sentence gives a meaning. If the T-sentences are understood prosententially, this stipulation can be put as follows: I presuppose that the tokening I interpret is a member of the set of anaphoric antecedents of the prosentence on the left-hand side of the T-sentence. Thus, anaphoric deflationism is compatible with the idea that T-sentences give the meanings of sentence types. They give the meanings of sentence types in that they give the meanings of 10 The primacy of tokenings over types in Davidsonian semantics is not limited to the first steps in radical interpretation (compare Davidson 1986). Because we always have to take into account the possibility of irony, metaphor, malapropisms and other phenomena, the theory we actually use to interpret an utterance is geared to the occasion (Davidson 1986, 441) it is geared to sentence tokenings at a given time for a given speaker. 10

11 all and only those sentence tokenings which are tokenings of that type. The same resources can also explain how prosententialism can accomodate the axiomatic character of Davidsonian semantics. For reasons independent of truth talk, radical interpreters start off with T-sentences about object language sentences as wholes. Which axioms about the internal structure of object language semtences are appropriate to unify this data is a further question. Interpreters have to stipulate how the object language sentences are structured, stipulate axioms in order to account for known T-sentences and test these axioms by deriving and testing hitherto unknown T-sentences. These axioms concern the structure of sentence types, in particular their phonetic and syntactic structure. Given that sentence types can be understood as sets of sentence tokenings, these axioms concern the structure of those tokenings which are members of the set in question. They specify which structural difference of an utterance makes which difference in meaning. On a prosentential reading of truth talk, such a set of axioms makes it possible to derive theorems including prosentences on the basis of the structure of the sentence tokenings these inherit their meanings from. 11 Thus, prosententialism can account for the axiomatic character of Davidsonian semantics. 3.2 Expression and Explanation However, Davidsonian semantics needs the T-sentences for a specific job, namely, explaining the meanings of the object language sentences. Davidson himself and the vast majority of his interpreters have taken this claim as follows: The theory explains the meanings of the object language sentences in terms of their truth-conditions. In short, it explains meaning in terms of 11 Of course, since some expressions are context-sensitive, interpreters need a set of axioms which includes contextual differences between utterances. But this does not threaten my point. In these cases, the set of anaphoric antecedents from which a prosentence inherits its meaning is picked out in terms of contextual features alongside structural ones. However, the general problem of context-sensitivity is beyond the scope of this paper. 11

12 truth. However, deflationism implies that truth is an expressive rather than a substantive concept, that the function of truth talk is wholly expressive, thus never explanatory. (Williams 1999, 547) Hence, it seems like no deflationist account of truth can provide what Davidsonian semantics requires. Let me take a step back. How should we understand the distinction between expression and explanation in the first place? Apparently, it is at the core of the debate on deflationism and semantics, but still it often remains implicit. In my view, the best way to make sense of this distinction is by analogy to the role of logical vocabulary in explanations like this one: 12 (12) An only child is a person without siblings. I take it that this is an explanation of the concept of an only child in terms of exactly two other concepts, the concepts of a person and a sibling. Of course, negation and conjunction also play a crucial, even indispensable role in (12). But their role is just to express the kind of explanatory relation between the concept of an only child on the one hand and the concepts of a person and a sibling on the other. (12) does not explain the concept of an only child in terms of the concepts of negation and conjunction and of a person and a sibling. Rather, it explains that concept in terms of the concepts of a person and a sibling, and it employs the expressive resources of negation and conjunction in order to accomplish this. I think that cases like this provide us with a sufficiently firm grip on the distinction between explanatory and expressive roles of concepts within explanations. The explanandum is explained in terms of all and only those elements which play explanatory roles. And the kind of explanatory relation between those elements is expressed by additional concepts which play expressive roles. These concepts express the form of the explanans, but they do not provide its foundation. They are not in terms of what the explanans 12 This analogy ties in with Brandom s broad notion of logical vocabulary, which explicitly includes semantic and intentional vocabulary (Brandom 1994 and 2008). However, my arguments here do not depend on such views on these additional topics. 12

13 purports to explain the explanandum, but how it purports to do so. 13 Let me now apply this distinction between explanatory and expressive roles of concepts within explanations to the T-sentences. Substantivists hold that (11) is an explanation of the meaning of Hunde bellen in terms of the proposition that dogs bark and in terms of truth. It explains the meaning of Hunde bellen in terms of the fact that Hunde bellen has the property truth just in case dogs bark. By contrast, the logical concept of bisubjunction only plays an expressive role. (11) is not an explanation of the meaning of Hunde bellen in terms of bisubjunction as well as the proposition that dogs bark and truth. Rather, it explains it in terms of the proposition that dogs bark and of truth, and it employs the expressive resources of the concept of bisubjunction in order to accomplish this. By contrast, deflationists hold that the concept of truth only does expressive work in the T-sentences. 14 Accordingly, (11) is an explanation of the meaning of Hunde bellen in terms of exactly one element, the proposition that dogs bark. Granted, the concept of truth plays an indispensable role in (11). But its role is comparable to that of bisubjunction: These concepts merely express the explanatory relation between the meaning of Hunde bellen and the proposition that dogs bark. (11) does not explain the meaning of Hunde bellen in terms of the concepts of truth and bisubjunction as well as the proposition that dogs bark. Rather, it explains it in terms of the proposition that dogs bark, and it employs the expressive resources of the concepts of truth and of bisubjunction in order to accomplish this. For everything I have said about Davidsonian semantics, the choice be- 13 Of course, this issue should be discussed much more thoroughly than I can provide in the context of this paper. For example, we should notice that the same explanatory and expressive resources can still lead to different explanations say, An only child is a nonperson with siblings instead of (12). Further, even though negation and conjunction play merely expressive roles in explanations like (12), they might very well play explanatory roles elsewhere, for example in defining the material conditional. By contrast, a defining feature of deflationism is that the concept of truth can never play an explanatory role. 14 Brandom (2002) grants a role for truth in such local explanations of content, but not in global explanations of contentfulness (compare section 4). 13

14 tween these two stories is entirely optional. The fact that the T-sentences explain the meanings of the object language sentences they quote leaves it open whether or not truth carries some of the explanatory weight. Thus, anaphoric deflationism is compatible with this idea. The whole explanatory burden can be put on the metalanguage proposition p. 3.3 Truth in Radical Interpretation Truth also plays important roles in the core of Davidsonian semantics, the theory of radical interpretation. Two of these issues stem from the principle of charity and have already been discussed by Williams (1999, 559ff.). However, the idea of triangulation brings up an important third point. First, it seems as if truth is involved in explaining logical form and logical consistency. But while logic makes use of the concept of truth, it is entirely neutral on the debate about theories of truth. To say that A and B are logically inconsistent because, in virtue of their logical form, they cannot both be true, is just to say that we can know that not-(a-and-b) solely in virtue of the logical forms of A and B. We can learn this, for example, by seeing that A implies not-b. Davidson relies on standard first-order logic in order to specify the logical forms of sentences. But nothing about first-order logic implies a commitment to a substantive conception of truth. It does not call for more than the expressive functions of that concept. Second, what an interpretee says should be interpreted as true whenever possible in order to understand them as rational and reasonable. This is crucial because the assumption of truth in canonical situations provides the much-needed entering wedge into the holistic circle of the interpretee s beliefs, desires and the meanings of her utterances. However, Davidson explicitly says that the standard for this assessment is not truth simpliciter, but what the interpreter believes, what she takes to be true. After all, interpreters can only work with what they have got. In trying to maximize rationality, the principle of charity aims at maximizing truth. But, in a catch-phrase, it aims 14

15 at maximizing truth in that it aims at maximizing agreement. Finally, however, triangulation brings a notion of truth back in which transcends mere agreement. Interpreters do not only have to make the beliefs they attribute to their interpretees agree with their own. They also have to conceive of both these and their own beliefs as capable of failing to be true against an objective standard of assessment. This is a fundamental realist distinction between what is true and what is merely believed. However, it is entirely innocent. Most importantly, it does not amount to a substantivist conception of truth, say, in terms of correspondence. Deflationists happily agree that there is a difference between merely believing that dogs bark and truly believing it. They simply add that truly believing that dogs bark boils down to believing that dogs bark while dogs actually bark. Again, the work done by the concept of truth is expressive rather than substantive. I conclude that what Davidsonian semantics needs from a conception of truth can be provided by anaphoric deflationism. 4 Circularity Ever since Michael Dummett (1959), deflationism and truth-conditional approaches to meaning have widely been pictured as incompatible because combining them would lead to an unacceptable form of circularity. In this section, I reconstruct the general pattern of such arguments and show why they cannot be applied to the present case (section 4.1). Then, I take a brief look at Claire Horisk s (2008) and Mark Lance s (1997) answers to the charge of circularity and argue that my strategy is superior to both (section 4.2). 15

16 4.1 Why the Circularity Objection Misfires What I take to be the most plausible version of this kind of argument can be expressed by the following schematic argument, the Circularity Objection: If deflationism is true, then the concept of truth is explained in terms of the linguistic mechanism M. 2. M can only be explained in terms of meaning. 3. For all concepts, if α is explained in terms of β, and β can only be explained in terms of γ, then γ cannot be explained in terms of α. 4. If deflationism is true, then meaning cannot be explained in terms of the concept of truth. Premise one is a general characterization of deflationism that follows from what I said above (section 2.1): Any variety of deflationism will claim, first, that the only thing to say about the concept of truth is to say how it functions, and second, that it functions according to some linguistic mechanism M. While disquotationalists claim that the functions of truth can be explained in terms of quoting and disquoting sentences, prosententialists hold that they can be explained in terms of anaphoric relations between sentence tokenings. Premise two is a second general claim about deflationism, but it does not directly follow from a general understanding of this position. However, this case is perfectly clear when it comes to prosententialism: Since the anaphoric relations appealed to are defined as relations of inheritance of meaning, explaining anaphoric relations requires appeal to meaning. Finally, premise three expresses a version of the principle that philosophical explanations of concepts should not be circular. It states that, if we explain, say, justice in terms of fairness and fairness in terms of moral equality, 15 This can be taken as an explication of what Horisk calls Explanatory Chicken-and- Egg Arguments (2008, 284ff.). Among many, John Collins (2002) argues like this. 16

17 we cannot rely on justice to explain moral equality. Intuitively, these concepts might then stand in a relationship of mutual illumination, but there would not be a full-blown explanation anymore. The present issue, however, concerns anaphoric deflationism and Davidsonian semantics. Let me fill in the first two premises of the schematic Circularity Objection and formulate a concrete argument about this case: 1. If anaphoric deflationism is true, then the concept of truth is explained in terms of anaphora. 2. Anaphora can only be explained in terms of meaning, namely in terms of inheritance of meaning. 3. For all concepts, if α is explained in terms of β, and β can only be explained in terms of γ, then γ cannot be explained in terms of α. 4. If anaphoric deflationism is true, then meaning cannot be explained in terms of the concept of truth. Some proponents of anaphoric deflationism use the conclusion of this argument in order to refute explanations of meaning in terms of truth. They affirm the antecedent and apply modus ponens. Among others, Brandom (1994, 2002) explicitly does so. Proponents of truth-conditional explanations of meaning, by contrast, can deny the consequent and apply modus tollens. One of those who explicitly do so is Donald Davidson (1990). In any case, if the argument is sound, these theories are incompatible. But, despite its widespread appeal, the Circularity Objection fails because it relies on a misconception of Davidsonian semantics (compare Williams 1999 and Kölbel 2001). The objection establishes an incompatibility between anaphoric deflationism and explanations of meaning in terms of truth. But, given the considerations of section 3, the commonplace characterization that Davidsonian semantics explains meaning in terms of truth is simply mistaken. 17

18 The fact that the concept of truth is used in the explanation does not entail that this concept does any of the explaining. The uses of is true can all be accounted for in terms of the expressive role of truth-talk. Davidsonian semantics is a truth-conditional approach to meaning, but it does not explain meaning in terms of truth. It explains the meanings of object language sentences in terms of the practice of interpreting them in a metalanguage. The explanatory work is done by the theory of radical interpretation alone. Thus, even if the Circularity Objection is sound, it simply misfires, because its conclusion does not apply to Davidsonian semantics Other Proposals Thus, the Circularity Objection fails because it falsely assumes that truth plays an explanatory role in Davidsonian semantics. Claire Horisk (2008) and Mark Lance (1997) have also argued against this objection. In this section, I will briefly show why Horisk s argument depends on the argument just given, and why Lance s argument cannot be applied to the present case. Horisk questions premise three of the Circularity Objection, what she calls the Priority Assumption. She distinguishes between two readings: On a metaphysical reading, circular explanations are unsound because they attempt to break down concepts into simpler and more fundamental concepts (Horisk 2008, 291) and because one concept cannot be both more and less simple than a second one. On an epistemological reading, a circular expla- 16 There might even be a second argument for this conclusion. Davidsonian semantics explains what object language sentences mean, but does this amount to an explanation of meaning in general? Davidson himself says that his theory illuminates the concept of meaning (1973, 1990, 1996), but it is not clear how this metaphor should be understood. Does it mean that meaning just is whatever the practice of interpretation reveals (Williams 1999, 553)? Or is the theory intended to deepen and clarify an independently available concept of meaning (compare Horisk 2008)? This is a complicated issue which I cannot fully account for in the present context. Maybe the best account of the metaphor of the illumination of meaning provides a further reason to reject the idea that meaning is explained in terms of truth. However, given that truth s role has turned out to be expressive rather than explanatory, there is sufficient reason to reject this idea anyway. 18

19 nation is unsound because the target concept turns out to be involved in its own understanding, making such an understanding impossible in the first place (compare Horisk 2008, 292). However, Horisk argues that both kinds of explanation do not have to take this reductive form and that Davidson s explanation of meaning does not do so. He spells out how meaning is connected to other core concepts such as truth, belief and desire. This does not reduce meaning to any of these, but still deepens our grasp of this concept. [H]e does not offer a reductive analysis, he offers a reciprocal analysis, that enriches our understanding of meaning by outlining the connections between the concepts of truth and meaning, and other concepts. (Horisk 2008, 291) But why exactly is a reciprocal explanation compatible with deflationism? And why exactly can reciprocal explanations be sound, even if they are circular? Horisk explicitly considers a version of this problem: [D]eflationists might worry that the account of truth will not be truly deflationary if they have any truck with a reciprocal analysis. But the worry may be unfounded. The issue must turn on the role of truth in the reciprocal account. If truth s role in the reciprocal analysis is restricted to logical and expressive functions, as Kölbel and Williams argue, then perhaps there is no cause for concern. (Horisk 2008, 291f.) But if this is true, Horisk s argument against the Circularity Objection depends on the soundness of the argument I gave above. The reason why a reciprocal explanation in terms of truth is sound even if circular is that truth only plays an expressive role. But if truth only plays an expressive role, there is no reason to speak of an explanation in terms of truth in the first place. This defeats the Circularity Objection directly, without Horisk s detour via reciprocal forms of explanation. I take it that, other things being equal, evading such an unnecessary detour is more elegant and less misleading. I shall now turn to Mark Lance (1997), who has presented an interesting general assessment of anaphoric deflationism including an argument against 19

20 the Circularity Objection. He starts with the observation that anaphoric deflationism validates the truth scheme (compare section 2.1): Truth Scheme φ is true just in case φ. Substantivists can be generally characterized as claiming that there is some substantial property T, say which applies to all and only truths. Realists might spell out T as correspondence to a fact, antirealists as maximal coherence, and so forth. The general substantivist position, however, is: Substantivism φ is true just in case φ has the substantial property T. Lance s point is simply this: Anaphoric deflationism does not contradict Substantivism. Rather, the Truth Scheme transforms Substantivism into: Substantivism* φ just in case φ has the substantial property T. Thus, he thinks that there is no incompatibility between Substantivism and anaphoric deflationism. By the same token, if truth was supposed to play any explanatory role in a theory of meaning or elsewhere, proponents of Substantivism can rely on the substantial property T instead. By cutting out the middle man, the concept of truth, all the explanatory work is done by what substantivists claim truth consists in, the substantial property T. For example, if I hold that truth consists in correspondence, I can leave out truth and explain meaning directly in terms of correspondence-conditions. But unfortunately, we cannot apply Lance s argument to Davidsonian semantics. It requires that the truth-conditional theory of meaning is supplemented with a version of Substantivism including a suitable truth-property T. But Davidson never subscribes to such a view and even explicitly rejects it in his latest work on truth (1990, 1996). For him, truth is an indefinable concept (1996, 265) and deflationists and substantivists alike are victims of the same ugly urge to define it (1996, 276). But if we cannot identify what truth consists in, no substantive property T, Lance s argument fails. For Davidson, truth cannot be a middle man one might also do without While Lance s argument fails when it comes to Davidsonian semantics, it may still 20

21 5 Compatibility Defended I will now discuss three further objections against the compatibility of Davidsonian semantics and anaphoric deflationism. 18 My responses to these worries will shed light on the modal status of the T-sentences (section 5.1) and the distinction between metalanguage and object language (sections 5.2 and 5.3), and establish the unity of the concept of truth accross languages (section 5.3). 5.1 Expressive Inadequacy Claire Horisk (2007) claims that a deflationary notion of truth cannot meet the expressive task it must be able to accomplish in Davidsonian semantics. Here, I will only discuss the version of this Expressive Inadequacy Argument, as she calls it herself, which concerns anaphoric deflationism (2007, 547ff.). The first step in this argument is to show that Davidsonian semantics requires a specific modal status for the T-sentences. They have to be contingent. Horisk develops this first step in the context of Max Kölbel s (2001) defense of the compatibility of deflationism and Davidsonian semantics, which also includes the biconditional doctrine, the first dogma of Davidsonian semantics. Roughly, this is the idea that T-sentences can give the meanings of object language sentences by specifying truth conditions, but that we cannot say what they mean directly. That is, we cannot use T-sentences such as (11) in order to derive what we might call M-sentences such as: (13) Hunde bellen means that dogs bark. Kölbel argues that if the T-sentences meet some independently important requirements, it is possible to infer M-sentences, after all (2001, 618ff.). On this succeed for other truth-conditional theories of meaning. In Löwenstein (2010), I argue that Lance relies on an incomplete understanding of substantivism, but that a more nuanced discussion of the relationship between substantivism and anaphoric deflationism reveals an important sense in which his argument is sound, after all. 18 I ignore a prominent objection by Bar-on, Horisk & Lycan (2001), the Determination Argument, since Douglas Patterson (2007) has already shown very clearly why it fails. 21

22 basis, Horisk can easily show that T-sentences must be contingent: Clearly, M-sentences are contingent. Thus, if we can sometimes correctly derive M- sentences from T-sentences, these T-sentences have to be contingent, too. Otherwise, these derivations would not be correct in the first place. Later, however, Horisk points out that this step does not depend on Kölbel s point: Whatever supported the claim that M-sentences have to be contingent directly supports the claim that T-sentences have to be contingent, too (compare Horisk 2007, 550). But she then leaves it to the reader to figure out how exactly the argument is supposed to generalize. However, I take it that she is correct, and that the argument can be put as follows: In Davidsonian semantics, T-sentences are empirical hypotheses aimed at making as much sense of the interpretee s linguistic behavior as possible. As empirical hypotheses, they constantly have to be reevaluated in the light of new observation and in the light of other T-sentences. Furthermore, they are used in order to explain the meanings of the object language sentences they quote. But the meanings of sentences are arbitrary, thus contingent. One and the same utterance whether oral or in writing might perfectly well mean something different from what it actually means. Thus, as empirical hypotheses, and as hypotheses about contingent matters (meanings), whether or not the T-sentences themselves are true has to be contingent, as well. Horisk is aware of the fact that the modal status of a language s semantic features is a matter of sophisticated philosophical discussion (2007, 550ff.). While I cannot discuss this topic here in more detail, I maintain the intuitively plausible claim that sentences might have meant something different from what they actually mean. 19 Those who reject it will thereby reject the first step of Horisk s argument. Since I grant it, I will have to show what 19 That the meaning of No bachelor is married is contingent entails that it is contingent that No bachelor is married is true. After all, it might have meant that cats can fly! But this is compatible with the idea that it is necessary that no bachelor is married. The existence of possible worlds where No bachelor is married means that cats can fly is compatible with the fact that no possible world contains a married bachelor. 22

23 is wrong with her second step, the idea that anaphoric deflationism violates the constraint that T-sentences are contingent (compare Horisk 2007, 549). Horisk argues that (11) is not merely a true T-sentence, but that it is necessarily true since the prosentence on its left-hand side inherits the meaning of the quoted sentence Hunde bellen. Hunde bellen means that dogs bark, hence the prosentence means that dogs bark, just like the sentence used on the right-hand side. Thus, both sides of the biconditional have the same meaning. And a claim that uses two synonymous expressions and asserts their equivalence is necessarily true. Since the left-hand side and the righthand side [..] have the same content [...], they are guaranteed to have the same truth-value in all possible worlds. (Horisk 2007, 549) Thus, prosententialism violates the constraint that the T-sentences have to be contingent. This may look like strong argument, but it fails eventually. In order to see why, recall the motivation of the constraint that T-sentences have to be contingent: They have to be contingent in that their truth depends on the meanings of the object language sentences they quote. But Horisk is wrong when she claims that anaphoric deflationism cannot account for this fact. Here is why. Intuitively, (11) is a true T-sentence because Hunde bellen means that dogs bark. This is contingent, since Hunde bellen might have meant that cats can fly. In this case, (11) would be false. Given that this is possible, the truth of (11) is contingent. The prosententialist version of this story is this: Since Hunde bellen means that dogs bark, the prosentence on the left hand side of (11) which inherits this meaning also means that dogs bark. But it is contingent that Hunde bellen means that dogs bark. Thus, it is equally contingent that the prosentence means that dogs bark. Hunde bellen might have meant that cats can fly. Then, the prosentence would also mean that cats can fly, and (11) would be false. Because it is possible that the prosentence means that cats can fly since it is possible that Hunde bellen means that cats can fly, the truth of (11) is contingent. Horisk fails to see this because she thinks that on a view like Brandom s, 23

24 the content of the left-hand side tracks the actual semantic features of the structually described sentence to which it refers (Horisk 2007, 550). On this view, it is necessary that Hunde bellen is true means that dogs bark, because its antecedent, in all possible worlds, is an utterance of Hunde bellen from the actual world. But this misconstrues anaphoric deflationism. The antecedent utterance of Hunde bellen is picked out structurally rather than semantically that is, in terms of which noise or which squibble was produced or perceived where and when. But it makes perfect sense to say that the same utterance exists in other possible worlds and that it can be picked out in the same way, 20 while its meaning varies across possible worlds. In sum, Horisk s worry is that prosententialism makes true T-sentences necessary because it makes them true in virtue of meaning. But there are possible worlds in which the meaning of the left hand side is not the same as the meaning of the right hand side. Where Hunde bellen means that cats can fly, (11) is false. In a sense, true T-sentences are indeed true in virtue of meaning, but not in virtue of their own meaning alone. They are true in virtue of the meaning of the object language sentence they quote. But this is not a problem, but what we wanted in the first place. 5.2 The Metalanguage Objection Let me now turn to Douglas Patterson s (2005) argument against the compatibility of Davidsonian semantics and deflationism what I call the Metalanguage Objection. 21 Patterson argues that what is at stake becomes clearer if we pay attention to the distinction between the object-language for which an account of truth and meaning is given, and the metalanguage in which this account is given (2005, 276). In particular, we need to distinguish between accounting for the object language truth-concept and accounting for 20 On identity criteria for utterances (tokenings) in terms of their structures, recall sections 2.1 and 3.1. In principle, these might even be cashed out in physical terms. 21 John Collins also hints at an argument along these lines (compare 2002, 520f.). 24

25 the metalanguage truth-concept (compare Patterson 2005, 279ff.). According to Patterson, deflationists claim that the meanings of object language sentences and applications of the concept of truth to these sentences are equivalent in what he calls basic meaning paradigmatically, in freestanding uses of those sentences (compare 2005, 277ff.). However, we must distinguish two versions of this idea. According to the first, the concept of truth in question belongs to the object language. Thus, what is at issue is the equivalence of Hunde bellen and Hunde bellen ist wahr, rather than Hunde bellen is true. This is object language deflationism, the claim that object language sentences s and attributions of object language truth to s are equivalent. The second idea in play is metalanguage deflationism, the claim that object language sentences s and attributions of metalanguage truth to s are equivalent. Here, the equivalence is between Hunde bellen and Hunde bellen is true rather than Hunde bellen ist wahr. Given this distinction, Patterson argues that while a truth-conditional theory of meaning is compatible with object language deflationism, it is incompatible with metalanguage deflationism (compare 2005, 280ff.). If, according to metalanguage deflationism, object language sentences s and attributions of metalanguage truth to s are equivalent, we can fix the meaning of our metalanguage truth-predicate is true in German on the basis of the meanings of the sentences of German. But then, this truth-predicate is of no help in explaining for the meanings of these object language sentences. We cannot at the same time use our metalanguage truth-predicate is true in German in such an explanation and claim that this predicate can be defined on the basis of the meanings of the object language sentences. Either is true in German explains the meanings of German sentences or the meanings of German sentences explain is true in German, but not both. It might be tempting to assimilate this argument to the Circularity Objection discussed above (section 4.1). But the Metalanguage Objection does not fall prey to the problems raised for the Circularity Objection. It acknowl- 25